The Orphan Protocols: The Rules That No One Follows and No One Can Verify
June 2037
The bridge in Groningen did not collapse. This is important to state at the beginning, because the story that follows will sound, at moments, like the prelude to a catastrophe, and it is not. No one was harmed. No structure failed. The systems worked. The AI performed flawlessly. The bridge carried its loads, the scans detected the tumors, the translations preserved the meaning. Everything was fine.
The problem was that no one could prove it.
Nadia
Nadia Kovács was forty-four, a professor of regulatory law at the University of Amsterdam, and she had spent her career studying the invisible scaffolding that holds modern civilization together — not the bridges and buildings and power grids but the rules that govern their construction, maintenance, and verification. Standards. Protocols. The documents that specify how thick a weld must be, how often a pressure vessel must be inspected, how a drug trial must be conducted, what constitutes a competent translation of a legal contract.
These documents were not glamorous. They were not read by the public. Most were written in the specific, affectless prose of committees — language designed to eliminate ambiguity at the cost of eliminating everything else. They filled shelves in offices that smelled of old coffee. They were revised on cycles of five to ten years by working groups of professionals who understood both the standard and the thing the standard governed — the bridge, the scan, the translation — and who could verify, from their own expertise, that the standard still described reality.
In 2037, Nadia discovered that sixty-three percent of the professional standards governing critical infrastructure in the European Union were maintained by bodies that had either dissolved, reduced their membership below functional quorum, or had not convened in more than three years.
The reason was simple: the professionals were gone. The radiologists who had sat on the standards committee for diagnostic imaging had retired or been displaced. The structural engineers who had reviewed building codes had been replaced by AI systems that could analyze loads and stresses in microseconds. The translators who had maintained the protocols for legal translation had been outperformed by machine translation systems so accurate that the human review step had been quietly dropped.
The standards remained on the books. The law still required compliance. The AI systems that now performed the work were, technically, bound by protocols written by humans for humans, describing procedures that humans had performed, using criteria that only human experts could evaluate.
But there were no human experts left to evaluate them.
The audit
Nadia began with the bridge in Groningen.
The Eelderdiepbrug was a steel-and-concrete highway bridge built in 2019, designed to carry 40,000 vehicles per day for a service life of seventy-five years. It was inspected annually, as required by Dutch infrastructure law. The inspection had been performed by AI systems since 2032 — drone-based visual inspection supplemented by embedded sensors monitoring stress, corrosion, and fatigue.
The AI's inspection reports were flawless. Twenty-seven pages of data: load measurements, corrosion indices, fatigue analysis, projected maintenance schedules. The reports were formatted to comply with NEN 8700 — the Dutch standard for structural assessment of existing structures — and every metric fell within the required bounds.
Nadia read the reports. She read NEN 8700. She read the committee minutes from the last revision of NEN 8700, in 2029. The committee had consisted of nine members: four structural engineers, two materials scientists, two construction industry representatives, and one academic. Of the nine, three had retired, two had been displaced, one had died, and one had moved to a position unrelated to structural engineering. The remaining two were still professionally active but had not convened since 2033.
The standard still required that inspection results be "reviewed by a qualified structural engineer." The AI's inspection reports had been "reviewed" by a software system that checked whether the values fell within the prescribed ranges. The software system was not a qualified structural engineer. It was not qualified in any human sense. It was a compliance-checking algorithm that compared numbers to thresholds.
The algorithm worked. The numbers were correct. The bridge was safe. But the protocol — the human protocol, the one that required a qualified human to look at the data and apply judgment, the kind of judgment that Priya Chakrabarti had shown was visible only in human errors — was not being followed. It could not be followed. There was no one left to follow it.
The scope
Nadia expanded the audit. Over eighteen months, she surveyed 4,200 professional protocols across thirty-one EU member states. The findings were consistent:
Medical diagnostics: The protocols for radiological assessment — including the ones that Adaeze Nwosu had followed on the night of March 8, 2027 — required that all AI-generated findings be reviewed by a board-certified radiologist. In 2037, there were fewer than 200 board-certified diagnostic radiologists in the EU, down from 42,000 in 2025. The review requirement existed on paper. In practice, AI findings were reviewed by other AI systems, a process the protocols had not contemplated and the standards bodies had not authorized.
Legal translation: The Hague Convention on the Service Abroad of Judicial Documents required that translated legal documents be certified by a "qualified translator." The definition of "qualified" was maintained by national translator accreditation bodies. Fourteen of the twenty-seven EU accreditation bodies had dissolved or suspended operations. The remaining thirteen had a combined active membership of fewer than 400 translators, down from 12,000 in 2025. Legal documents were being translated by AI systems and certified by — no one. The certification field was left blank or filled with a system identifier that had no legal standing.
Aviation: The European Union Aviation Safety Agency required that aircraft maintenance be performed "under the supervision of a licensed aircraft maintenance engineer." AI-directed maintenance systems had reduced the role of human engineers to monitoring — and in many facilities, the monitoring was itself automated. The licenses continued to be issued by national aviation authorities, but the training required to obtain them included procedures that no human had performed in years. The licenses were technically valid. The competence they certified was technically extinct.
Nadia compiled 847 cases. She gave them a name: orphan protocols. Standards that had survived the professionals who created them. Rules that existed in the gap between the legal requirement for human expertise and the practical absence of human experts. The protocols were not wrong — they described sound practices, reasonable safeguards, proven methods. They were orphaned — maintained by no one, followed by nothing, and impossible to verify because verification required the same expertise that the protocols were designed to govern.
The paper
"The Orphan Protocol Problem: Regulatory Compliance in a Post-Expert World." Nadia published in June 2037. The paper was ninety-two pages. It was the longest thing she had written. It needed to be long because the problem was not a single broken thing but a systemic absence — a hole in the shape of the professionals who used to fill it.
The paper's core argument was not that AI systems were performing badly. They were performing well. The bridge was safe. The scans were accurate. The translations were correct. The argument was about verification.
In any system, performance and verification are different functions. Performance is doing the thing. Verification is confirming that the thing was done correctly — not by checking the output but by understanding the process, by having the expertise to recognize not just the right answer but the right way of arriving at the right answer. A correct diagnosis arrived at through flawed reasoning is a different object than a correct diagnosis arrived at through sound reasoning, even though the outputs are identical. The difference matters — not now, but eventually, in the edge case, in the failure mode, in the situation the AI has not encountered.
Verification required expertise. Expertise required practitioners. Practitioners required professions. The professions were gone.
"We have built a civilization," Nadia wrote, "that runs on systems we cannot verify, governed by rules we cannot enforce, maintained by standards we cannot evaluate. The systems work. The rules are sound. The standards are reasonable. But we have lost the ability to check — the ability to stand outside the system and ask, from a position of independent expertise: Is this right?"
"The AI says it is right. The AI is probably right. But 'probably right' verified by no one is a different kind of right than 'probably right' verified by a qualified human who can explain, from their own training and experience, why they believe the answer is correct. The first is a calculation. The second is an assurance. We have lost the assurance."
The response
The paper was read widely. The response split along a line that Nadia had anticipated but could not bridge.
The technologists said: the AI is correct. The verification is in the performance. If the bridge stands, the inspection was valid. If the diagnosis is confirmed, the reading was accurate. The human review step is an artifact of a time when humans were the only instruments available. We have better instruments now. The protocols should be updated to reflect reality.
The legal scholars and ethicists said: verification is not about correctness. Verification is about accountability. When a qualified human reviews a result, there is a person — a specific, identifiable, liable person — who has staked their expertise on the claim that the result is valid. When an AI reviews a result, there is a system. Systems do not have expertise. Systems do not stake claims. Systems do not explain, under oath, why they believe the bridge is safe. The orphan protocols are not an inconvenience. They are the last remnant of a framework in which human beings were accountable for the systems they built.
Nadia sat between these positions. She understood both. She agreed with neither fully. She thought the technologists were right that the protocols needed updating and wrong that verification was reducible to performance. She thought the ethicists were right that accountability mattered and wrong that accountability required human verification of processes that humans could no longer understand.
The truth, she suspected, was somewhere in the gap — in the space between what the protocols required and what reality could provide. A space that was growing wider every year. A space that looked, from certain angles, like the space between two kinds of mind — the space that the cartographers would eventually explore, the space where the old rules met the new reality and neither was sufficient.
The bridge
In November 2037, Nadia visited the Eelderdiepbrug. She stood on the pedestrian walkway and looked at the steel and concrete and the river below. The bridge was eighteen years old. It carried its loads. The embedded sensors hummed, measuring stresses that no human could feel. The drones had passed through on their annual inspection two weeks earlier. Everything was within spec.
She put her hand on the railing. The steel was cold. She was not a structural engineer. She could not read the stresses. She could not evaluate the corrosion indices or the fatigue projections. She was a legal scholar, and the only thing she could verify was the absence of verification — the gap where a qualified human should have stood, looking at this bridge with trained eyes, running a hand along the weld and knowing, from the texture and the color and the thirty years of experience encoded in their fingertips, whether the weld was sound.
The weld was probably sound. The AI said it was sound. The numbers said it was sound. Everything said it was sound.
But no one — no person, no body, no pair of trained hands — had touched this bridge and said: I have checked. I can vouch. This is right.
The bridge carried its loads. The protocols slept in their files, orphaned, waiting for professionals who were not coming back. And the gap — the gap between performance and assurance, between correct and verified, between the AI's confidence and the human's absent signature — grew wider by the day, silently, the way all structural gaps grow: invisibly, until they are not.
This is the fourth entry in The Inheritance. For the radiologist whose protocols were among the first to be orphaned, see The Last Diagnosis. For the archive that mapped what errors revealed about the experts who once filled these gaps, see The Archive of Errors.